Pedram Hatef — Perpetual-futures bot auditor

/ what i catch

Perp bots rarely blow up loudly. They bleed: a stop-loss placed a few ticks off, an indexer read with no retry, a safety flag that latches on and silently halts trading. I read execution code for exactly these failure modes.

Fee-blind TP/SL placementTargets and stops computed without accounting for fees or tick size — so the real exit lands somewhere the risk model never assumed.
No-retry indexer readsPosition and fill data pulled from the indexer with no retry or backoff. One flaky read and the bot acts on stale or missing state.
Sticky safety flagsLocks and circuit breakers that latch on after an incomplete trade and never release — quietly halting trading on a market with no auto-recovery.
Orphaned conditional ordersA take-profit or stop left live on the book after the position closes, ready to fire as a fresh, unintended trade.
Decimal & precision errorsPrices and sizes that don't round to the instrument's tick/step — rejected outright, or worse, accepted at the wrong level.
Order-lifecycle racesEntry, TP, SL, and cancel paths that can interleave under load and leave the account in a state nobody designed for.

/ how it works

Start free. Most bots have at least one money-losing bug I can point to inside 30 minutes.

STEP 01Free

30-minute review

I read your execution code and flag the top risks I see, live. No charge, no obligation — you walk away with something useful either way.

STEP 02$1,500 · ~1 week

Fixed-scope audit

A full pass over order placement, position lifecycle, and recovery paths. You get a written report ranking every issue by money-at-risk, plus a one-hour debrief call.

STEP 03$100 / hr

Ongoing engineering

Implement the fixes, add the missing guardrails, or build out new execution logic — on call as a second set of hands that already knows your code.

/ incident reports

Two real bugs I traced and fixed in a production multi-asset perp bot on dYdX v4. Both are the kind that cost money silently, with nothing in the logs that screams.

INCIDENT 01 RESOLVED Mispriced TP/SL left a leveraged position unprotected

Situation

The bot computed take-profit and stop-loss triggers from entry price, target PnL, and position size, then submitted them as conditional orders on dYdX v4.

Complication

The trigger math didn't round to the instrument's tick size. A non-conforming price gets rejected — or lands a few ticks off intended. On leverage, a stop sitting at the wrong level turns a capped loss into an open-ended one, and nobody notices until the market moves.

Action

Traced the calculation, corrected the formula (for a short: trigger = entry − targetPnL ÷ size), and enforced tick-size rounding plus pre-submission validation so no malformed or mis-leveled trigger reaches the exchange.

Outcome

Stops and targets now submit at correct, tick-conformant levels every time. The silent "unprotected position" failure mode is gone.

INCIDENT 02 RESOLVED Stale in-flight lock froze trading on a market

Situation

To stop two trades colliding on the same market, the bot held an in-memory lock per symbol — one trade in flight at a time.

Complication

When a trade lifecycle didn't finish cleanly, the lock stayed stuck on. Every later signal hit a 409 already-in-flight, the app reported the backend unreachable, and the bot quietly stopped trading that market until someone restarted it by hand — a silent outage with no auto-recovery.

Action

Diagnosed the stuck lock as an incomplete-lifecycle artifact and added a 30-second lock timeout, so a stale lock expires on its own — turning a manual-recovery halt into a self-healing one.

Outcome

A hung trade no longer freezes a market. The system recovers within seconds, unattended.

/ open source

perps-bot-healthcheck

I packaged these checks into a Python static analyzer that flags fee-blind TP/SL, no-retry indexer reads, sticky safety flags, and orphaned-order risk across a perp-bot codebase. Read the code, run it on your own bot.

github.com/pedramhatef/perps-bot-healthcheck →

/ contact

Running a perp bot and want a second set of eyes on the execution code? The 30-minute review is free — bring your worst-behaving market.

contact@pedramhatef.dev github.com/pedramhatef →